Background - In France, when presenting the results of a random sample survey, the non-response rate is often cited as the sole criterion of quality. It is generally believed that non-participation can introduce bias into the estimation of indicators derived from the collected data. However, the significant variability in non-response rates reported in surveys that use identical methods raises questions not only about the very method of constructing these indicators but also about their significance as a measure of quality. Methods - This issue will be addressed through a detailed analysis of the address-based consumption database from two surveys conducted in France using an identical data collection protocol but with very different non-response rates, ranging from single to double. Results - The two surveys compared indeed show refusal rates of 35.6% for the Nicolle survey and 18.3% for the KABP AIDS survey. However, the number of phone numbers required to conduct an interview is higher for the KABP survey: 2.8 addresses were needed to complete an interview, compared to 2.1 addresses for the Nicolle survey. If we now consider not only the refusal rates—which remain different—but also other factors, we see that these two surveys have, overall, an identical participation rate: 55.1% for the Nicolle survey and 56.2% for the KABP AIDS survey. This result, which is not apparent if one relies solely on the refusal rate, stems from a higher number of telephone numbers coded as unusable in the KABP study. Conclusion - It appears essential to move beyond presenting the refusal rate as the sole qualitative criterion for the fieldwork of a random sample survey in public health, using, for example, the work of AAPOR. An alternative, which we believe would be easier to use, would be to consider a set of three rates whose sum would total 100% (participation, refusal, and failure to complete). This would provide more meaningful and less misleading information than the refusal rate.